Slashdot Mirror
The Mother of All CPU Charts
||Plazm|| writes "Tom's Hardware has an entertaining read on the latest offerings from processor makers Intel and AMD. Not only does it contain a plethora of benchmarks on the latest Dual core CPU's, but it also includes benchmarks from over 60 other legacy processors. Better yet, they let the benchmarks speak for themselves and let you draw your own conclusions. You may want to fill up your 44oz mug before sifting through this one, though."
Now AMD doesn't even make chipsets and their mobile offering is still quite the joke in the face of the Pentium M.
From what I've heard, the new AMD mobile chip ("Turon" I think) has pretty much caught up with the Pentium M, and is far better than the old AMD mobile junk.
The Pentium M has a much bigger L2 cache, but the Turon has AMD's typically better memory interface, amd64 mode, etc.; the reviews I saw seemed to basically call it a wash (i.e., the results can go either way depending on which benchmark you use). In any case, AMD's clearly back in the mobile game.
We live, as we dream -- alone....
Processor throughput has increased tremendously. Clock speed has not increased. Issue widths are wider. Larger, faster, and more effective caches are being used, in addition to the introduction of trace caches. Branch prediction continues to gets better, along with speculation techniques. More physical registers and larger lookahead windows allow modern CPUs to pull more parallelism out of single threaded programs than ever before.
Features like hyper-threading and dual cores give a much greater system wide speedup than simply raising the clock rate, and avoid all the problems of power consumption. Even on single thread performance, having another core to run the OS, so your not constantly context switching, can make a differance.
Reading this article made me sick, because they equate speed with clock rate. This is patently false, as the last two years of computer architecture have shown us.
-- the computer doesn't want any beer, no matter how much you think it does. NEVER, EVER feed your computer beer.
http://www.mobilityguru.com/2005/08/30/the_turion
http://www.mobilityguru.com/2005/09/06/the_turion
It gives a good handle on the AMD chips for laptops. All in all it holds it own with the Pentium M, where the Pentium M has a good lead is in power saving.
Also the AMD flagship laptop chip is 64bit so you would see a big jump in performace if you were to run a 64bit OS/Apps as you would except.
I like the look of the AMD chips over all and feel that Intel has drop the ball on the x86 market and put the eggs in the Intamin basket. And that ship is going down faster than Kate Mose can do a line :P
"The most dangerous creation of any society is that man who has nothing to lose." - James Baldwin, American author
Because the benchmarks are pretty much single threaded, so only one core is actually being used, so on these particular benchmarks a dual core CPU won't show any benefit, it might even suffer a little, if the fact of it being dual core allowed a small background task that might have otherewise stalled to run and was therefore fighting for the cache.
Heh, you call that a list?
SPECcpu beats this hands down. THG is great and all, but SPEC are a non-profit organization *dedicated* to measuring the performance of computing systems. Believe me when I say their "CPU 2000" benchmark is not only the standard benchmark, but the *best* standard benchmark out there. It's cross-platform: Windows, Linux, HP-UX, AIX, whatever: you name it, it's been tested. It's cross-compiler: GCC, Intel ICC, AMD/Pathscale, IBM xlC, they're all here.
Here's the list. It's big.
Enjoy.
I thought I remembered this article from a few days ago. It seems that even Dell is finally starting to see the light.
Could it be that Intel's days as a CPU manufacturer are numbered?
I put together this web-scraping chart comparing price/performance of available x86-compatible CPU families on http://dclug.tux.org/cpu.png. The daily-run script collects web-advertised prices, and displays them as a series of lines showing price-vs.-nominal clockspeed data within the CPU sub-families.
Note the logarithmic scale of the Y (price in US$) axis---in linear scale it's easier to see the knee in the curve, where additional speed increments begin to cost disproportionately more, but the linear graph was much less readable.
The method used is a hacked-together heuristic, easily fooled by vendors' ever inventive approaches to reporting the parameters of the CPUs they are selling. Still, it's a good visual aid showing how various CPU families are positioned
with respect to each other.